Method and apparatus for repairing GPU video memory access based on active error detection

An apparatus to repair access to a graphics processing unit (GPU) video memory based on active error detection, the apparatus comprising:

• a regular detection module configured to allocate a video memory test space on which data read and write detection is performed, configured to detect, with a predetermined frequency, whether read and write of data in the video memory test space is normal, wherein the video memory test space is two times or four times a width of video memory data, and configured to trigger, with a predetermined frequency, the regular detection module to perform the data read and write detection on the video memory test space, wherein the detecting comprises:
• randomly writing a group of data to the video memory test space, the group of data being written data, and
• reading out the written data as read data and performing correctness verification on the read data, such that when the read data is consistent with the written data, correctness verification indicates that the read and write of data is normal, and when the read data is inconsistent with the written data, correctness verification indicates that the data read and write detection fails;
• an instruction trigger module configured to transmit a trigger instruction when the correctness verification indicates that the data read and write detection on the video memory test space fails;
• a GPU desktop driving module, when receiving the trigger instruction, configured to perform repair operations of access to the GPU video memory comprising:
• suspending access operations to the GPU video memory,
• performing a desktop refresh operation after restoring an access to a video memory interface to normal, and restoring all access operations to the GPU video memory, comprising:
• setting a full screen update mark to ensure that an entire desktop picture is updated in next desktop updating;
• restoring a graphic desktop drawing operation process to normal, and restoring the data read and write detection on the video memory test space;
• and
• a GPU kernel driving module, when receiving the trigger instruction, configured to perform repair operations of access to the GPU video memory comprising:
• suspending access operations of all hardware modules to the GPU video memory,
• re-calibrating timing of the video memory interface by resetting and re-initializing a video memory controller, and
• restoring the access operations of all the hardware modules to the GPU video memory and restoring the access to the video memory interface to normal.